[macOS][26.810.52044] Ctrl+Shift+M opens /model submenu instead of full model/reasoning/Fast picker
What version of the Codex App are you using (From “About Codex” dialog)?
26.810.52044 (build 6662; bundled codex-cli 0.148.0-alpha.9)
What subscription do you have?
ChatGPT subscription; this does not appear to be plan-specific.
What platform is your computer?
Darwin 25.5.0 arm64 arm
What issue are you seeing?
In the latest macOS desktop build, Ctrl+Shift+M opens a lightweight, model-only selector that appears to be the same submenu opened by typing /model.
Previously, this shortcut opened the full composer picker available from the lower-right control. That richer picker exposes the model, reasoning effort, and Fast/service-tier controls together.
The shortcut therefore no longer provides keyboard access to reasoning effort or Fast mode, even though the full picker is still present and usable with the mouse. This is a desktop UX regression rather than a backend model-selection failure.
What steps can reproduce the bug?
- Open a local Codex task in the macOS desktop app.
- Click the model control at the lower-right of the composer.
- Observe the full picker with model, reasoning effort, and Fast/service-tier controls.
- Close the picker.
- Press
Ctrl+Shift+M. - Observe a simple model-only list.
- Close it, type
/model, and observe the same model-only submenu.
This reproduces while the task is idle and does not depend on prompt content or conversation length.
What is the expected behavior?
Ctrl+Shift+M should open the full composer model picker—the same rich surface opened from the lower-right composer control—so model, reasoning effort, and Fast/service-tier settings remain keyboard-accessible.
If the model-only slash-command submenu needs its own shortcut, it should use a separate command rather than replacing the existing composer.openModelPicker behavior.
Additional information
A sanitized inspection of the installed app bundle at /Applications/ChatGPT.app/Contents/Resources/app.asar supports the observed behavior:
- The default keybinding still maps
Ctrl+Shift+Mtocomposer.openModelPicker. - The only literal handler registration for that command in the shipped webview bundle opens the slash-command
modelview. - The richer picker remains in the same bundle, including its simple/advanced views, reasoning slider, model rows, and Fast/service-tier handling, but it no longer owns the literal
composer.openModelPickerhandler.
This suggests a frontend command-routing regression, but the bundle evidence alone does not establish the intended product behavior or exact source commit.
As supporting—not definitive—comparison evidence, an installed 26.803.61601 VS Code frontend bundle registers composer.openModelPicker directly to its rich picker open function. Because that is a different application target, it should not be treated as an exact desktop regression bisection.
I searched existing issues and did not find an exact report. Related but distinct reports:
- #26819 tracks keyboard shortcuts for model/reasoning selection and received a recent comment about lost reasoning-effort keyboard access.
- #31968 documents the full Advanced model/reasoning picker.
- #22279 concerns the same shortcut failing under non-English input sources, not the wrong picker opening.
4 Comments
Reproduced on the current macOS desktop app. This is a meaningful keyboard-workflow regression: I often keep the same model but switch reasoning effort between lightweight work and deeper debugging/planning. Please restore the full composer picker for this shortcut, or provide a dedicated reasoning-effort shortcut.
I'm an API user and ctrl+shift+m just stopped working for me completely this week. I don't even see model-only selector mentioned above. I also don't have
/modelas a command in my codex app.This exists @Geraldxm. However this is still a bug and I'd like to get this reverted/fixed.
<img width="562" height="375" alt="Image" src="https://github.com/user-attachments/assets/37aca33c-2ad1-45a3-8103-4c4442f3da68" />
This issue has been opened for days but still receives no attention from openai's team.
After several days of use of the new behavior, paired with my personalized
ctrl-shift-, and .to decrease / increase reasoning effort, I think this change, despite of being unintentional, works pretty well if the person wants full fledged control over model + effort.A better fix maybe separate shortcut for the model picker and widget in composer. Then we have both world. Or at least expose an configuration entry for widget mode.
all in all, I will close it for now. But still wish a polished fix for this as I have said above.